Getting Started

Installation

Alibi Detect can be installed from PyPI or conda-forge by following the instructions below.

Install via PyPI

Alibi Detect can be installed from PyPI with pip. We provide optional dependency buckets for several modules that are large or sometimes tricky to install. Many detectors are supported out of the box with the default install but some detectors require a specific optional dependency installation to use. For instance, the OutlierProphet detector requires the prophet installation. Other detectors have a choice of backend. For instance, the LSDDDrift detector has a choice of tensorflow or pytorch backends. The tabs below list the full set of detector functionality provided by each optional dependency.

Default installation.

pip install alibi-detect

The default installation provides out the box support for the following detectors:

If you are unsure which detector to use, or wish to have access to as many as possible the recommended installation is:

pip install alibi-detect[tensorflow,prophet]

If you would rather use pytorch backends then you can use:

pip install alibi-detect[torch,prophet]

However, the following detectors do not have pytorch backend support:

Alternatively you can install all the dependencies using (this will include both tensorflow and pytorch):

pip install alibi-detect[all]

Note

If you wish to use the GPU version of PyTorch, or are installing on Windows, it is recommended to install and test PyTorch prior to installing alibi-detect.

Note

If using torch version 2.0.0 or 2.0.1 along with some versions of tensorflow you may experience hanging depending on the order you import each of these libraries. This is fixed in torch 2.1.0 onwards.

Installation with PyTorch backend.

pip install alibi-detect[torch]

The PyTorch installation is required to use the PyTorch backend for the following detectors:

Note

If you wish to use the GPU version of PyTorch, or are installing on Windows, it is recommended to install and test PyTorch prior to installing alibi-detect.

Note

If using torch version 2.0.0 or 2.0.1 along with some versions of tensorflow you may experience hanging depending on the order you import each of these libraries. This is fixed in torch 2.1.0 onwards.

Installation with TensorFlow backend.

pip install alibi-detect[tensorflow]

The TensorFlow installation is required to use the TensorFlow backend for the following detectors:

The TensorFlow installation is required to use the following detectors:

Installation with KeOps backend.

pip install alibi-detect[keops]

The KeOps installation is required to use the KeOps backend for the following detectors:

Note

KeOps requires a C++ compiler compatible with std=c++11, for example g++ >=7 or clang++ >=8, and a Cuda toolkit installation. For more detailed version requirements and testing instructions for KeOps, see the KeOps docs. Currently, the KeOps backend is only officially supported on Linux.

Installation with Prophet support.

pip install alibi-detect[prophet]

Provides support for the OutlierProphet time series outlier detector.

Install via conda-forge

  • To install the conda-forge version it is recommended to use mamba, which can be installed to the base conda enviroment with:

conda install mamba -n base -c conda-forge

  • mamba can then be used to install alibi-detect in a conda enviroment:

mamba install -c conda-forge alibi-detect

Features

Alibi Detect is an open source Python library focused on outlier, adversarial and drift detection. The package aims to cover both online and offline detectors for tabular data, text, images and time series. TensorFlow, PyTorch and (where applicable) KeOps backends are supported for drift detection. Alibi-Detect does not install these as default. See installation options for more details.

To get a list of respectively the latest outlier, adversarial and drift detection algorithms, you can type:

import alibi_detect

# View all the Outlier Detection (od) algorithms available
alibi_detect.od.__all__

['OutlierAEGMM',
 'IForest',
 'Mahalanobis',
 'OutlierAE',
 'OutlierVAE',
 'OutlierVAEGMM',
 'OutlierProphet',
 'OutlierSeq2Seq',
 'SpectralResidual',
 'LLR']

# View all the Adversarial Detection (ad) algorithms available
alibi_detect.ad.__all__

['AdversarialAE',
'ModelDistillation']

# View all the Concept Drift (cd) detection algorithms available
alibi_detect.cd.__all__

['ChiSquareDrift',
 'ClassifierDrift',
 'ClassifierUncertaintyDrift',
 'ContextMMDDrift',
 'CVMDrift',
 'FETDrift',
 'KSDrift',
 'LearnedKernelDrift',
 'LSDDDrift',
 'LSDDDriftOnline',
 'MMDDrift',
 'MMDDriftOnline',
 'RegressorUncertaintyDrift',
 'SpotTheDiffDrift',
 'TabularDrift']

Summary tables highlighting the practical use cases for all the algorithms can be found here.

For detailed information on the outlier detectors:

Similar for adversarial detection:

And data drift:

Basic Usage

We will use the VAE outlier detector to illustrate the usage of outlier and adversarial detectors in alibi-detect.

First, we import the detector:

from alibi_detect.od import OutlierVAE

Then we initialize it by passing it the necessary arguments:

od = OutlierVAE(
    threshold=0.1,
    encoder_net=encoder_net,
    decoder_net=decoder_net,
    latent_dim=1024
)

Some detectors require an additional .fit step using training data:

od.fit(X_train)

The detectors can be saved or loaded as described in Saving and loading. Finally, we can make predictions on test data and detect outliers or adversarial examples.

preds = od.predict(X_test)

The predictions are returned in a dictionary with as keys meta and data. meta contains the detector’s metadata while data is in itself a dictionary with the actual predictions (and other relevant values). It has either is_outlier, is_adversarial or is_drift (filled with 0’s and 1’s) as well as optional instance_score, feature_score or p_value as keys with numpy arrays as values.

The exact details will vary slightly from method to method, so we encourage the reader to become familiar with the types of algorithms supported in alibi-detect.